Date:	Wed, 14 Aug 2013 13:08:34 -0700
From:	Sudeep Dutt <sudeep.dutt@...el.com>
To:	Greg Kroah-Hartman <gregkh@...uxfoundation.org>
Cc:	Arnd Bergmann <arnd@...db.de>,
	Rusty Russell <rusty@...tcorp.com.au>,
	"Michael S. Tsirkin" <mst@...hat.com>,
	Rob Landley <rob@...dley.net>, linux-kernel@...r.kernel.org,
	virtualization@...ts.linux-foundation.org,
	linux-doc@...r.kernel.org, AsiasHeasias@...hat.com,
	Nikhil Rao <nikhil.rao@...el.com>,
	Ashutosh Dixit <ashutosh.dixit@...el.com>,
	Caz Yokoyama <Caz.Yokoyama@...el.com>,
	Dasaratharaman Chandramouli 
	<dasaratharaman.chandramouli@...el.com>,
	Harshavardhan R Kharche <harshavardhan.r.kharche@...el.com>,
	"Yaozu (Eddie) Dong" <eddie.dong@...el.com>,
	Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@...el.com>,
	Sudeep Dutt <sudeep.dutt@...el.com>
Subject: Re: [PATCH v2 1/7] Intel MIC Host Driver for X100 family.

On Mon, 2013-08-12 at 16:06 -0700, Greg Kroah-Hartman wrote:
> On Wed, Aug 07, 2013 at 08:04:07PM -0700, Sudeep Dutt wrote:
> > +/**
> > + * struct mic_device -  MIC device information for each card.
> > + *
> > + * @name: Unique name for this MIC device.
> > + * @mmio: MMIO bar information.
> > + * @pdev: The PCI device structure.
> > + * @family: The MIC family to which this device belongs.
> > + * @ops: MIC HW specific operations.
> > + * @id: The unique device id for this MIC device.
> > + * @stepping: Stepping ID.
> > + * @attr_group: Sysfs attribute group.
> > + * @sdev: Device for sysfs entries.
> > + * @aper: Aperture bar information.
> > + */
> > +struct mic_device {
> > +	char name[20];
> 
> The name can be in the struct device (it should be the same, right?)
> 

The name field is redundant and will be deleted.

> > +	struct mic_mw mmio;
> > +	struct pci_dev *pdev;
> 
> Isn't this just the parent of the device?  Do you really need this?
> 

pdev is redundant as well and will be deleted.

> > +	enum mic_hw_family family;
> > +	struct mic_hw_ops *ops;
> > +	int id;
> > +	enum mic_stepping stepping;
> > +	struct attribute_group attr_group;
> 
> Shouldn't this be a pointer to a list of groups?
> 

We have a single attribute group. However a list of groups is scalable
and also required by device_create_with_groups(..) so we have changed
this to a list of groups.

> > +	struct device *sdev;
> 
> Shouldn't this be embedded inside here, instead of a pointer?
> 

We are using device_create(..) (or device_create_with_groups(..) now)
which returns a pointer to the device.

> > +	struct mic_mw aper;
> > +};
> 
> 
> > +/**
> > + * struct mic_info -  Global information about all MIC devices.
> > + *
> > + * @next_id: Next available MIC device id.
> > + * @mic_class: Class of MIC devices for sysfs accessibility.
> > + * @mdev_id: Base device node number.
> > + */
> > +struct mic_info {
> > +	int next_id;
> 
> Please use the idr interface, don't roll your own, odds are you got it
> wrong, and I don't want to have to debug it :(
> 

Agreed, the IDA interface works nicely here.

> > +	struct class *mic_class;
> 
> Isn't this a global symbol that you have (or static symbol).  There
> should never be more than one "class" around for these devices.
> 

There is only one class for these devices stored in g_mic static symbol,
as you noticed below. We will improve the documentation here but please
let us know if you were suggesting other changes.

> > +	dev_t mdev_id;
> > +};
> > +
> > +/* g_mic - Global information about all MIC devices. */
> > +static struct mic_info g_mic;
> 
> See, one class :)
> 
> > +/**
> > + * mic_probe - Device Initialization Routine
> > + *
> > + * @pdev: PCI device structure
> > + * @ent: entry in mic_pci_tbl
> > + *
> > + * returns 0 on success, < 0 on failure.
> > + */
> > +static int __init mic_probe(struct pci_dev *pdev,
> > +		const struct pci_device_id *ent)
> > +{
> > +	int rc;
> > +	struct mic_device *mdev;
> > +	char name[20];
> > +
> > +	rc = g_mic.next_id++;
> 
> No locking, please use the idr interface...
> 

Agreed, the IDA interface works nicely here.

> > +
> > +	snprintf(name, sizeof(name), "mic%d", rc);
> > +	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
> > +	if (!mdev) {
> > +		rc = -ENOMEM;
> > +		dev_err(&pdev->dev, "dev kmalloc failed rc %d\n", rc);
> > +		goto dec_num_dev;
> > +	}
> > +	strncpy(mdev->name, name, sizeof(name));
> > +	mdev->id = rc;
> > +
> > +	mic_device_init(mdev, pdev);
> > +
> > +	rc = pci_enable_device(pdev);
> > +	if (rc) {
> > +		dev_err(&pdev->dev, "failed to enable pci device.\n");
> > +		goto free_device;
> > +	}
> > +
> > +	pci_set_master(pdev);
> > +
> > +	rc = pci_request_regions(pdev, mic_driver_name);
> > +	if (rc) {
> > +		dev_err(&pdev->dev, "failed to get pci regions.\n");
> > +		goto disable_device;
> > +	}
> > +
> > +	rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
> > +	if (rc) {
> > +		dev_err(&pdev->dev, "Cannot set DMA mask\n");
> > +		goto release_regions;
> > +	}
> > +
> > +	mdev->mmio.pa = pci_resource_start(pdev, mdev->ops->mmio_bar);
> > +	mdev->mmio.len = pci_resource_len(pdev, mdev->ops->mmio_bar);
> > +	mdev->mmio.va = pci_ioremap_bar(pdev, mdev->ops->mmio_bar);
> > +	if (!mdev->mmio.va) {
> > +		dev_err(&pdev->dev, "Cannot remap MMIO BAR\n");
> > +		rc = -EIO;
> > +		goto release_regions;
> > +	}
> > +
> > +	mdev->aper.pa = pci_resource_start(pdev, mdev->ops->aper_bar);
> > +	mdev->aper.len = pci_resource_len(pdev, mdev->ops->aper_bar);
> > +	mdev->aper.va = ioremap_wc(mdev->aper.pa, mdev->aper.len);
> > +	if (!mdev->aper.va) {
> > +		dev_err(&pdev->dev, "Cannot remap Aperture BAR\n");
> > +		rc = -EIO;
> > +		goto unmap_mmio;
> > +	}
> > +
> > +	mdev->ops->init(mdev);
> > +
> > +	pci_set_drvdata(pdev, mdev);
> > +
> > +	mdev->sdev = device_create(g_mic.mic_class, &pdev->dev,
> > +		MKDEV(MAJOR(g_mic.mdev_id), mdev->id), NULL, "%s", mdev->name);
> > +	if (IS_ERR(mdev->sdev)) {
> > +		rc = PTR_ERR(mdev->sdev);
> > +		dev_err(&pdev->dev, "device_create failed rc %d\n", rc);
> > +		goto unmap_aper;
> > +	}
> > +
> > +	rc = sysfs_create_group(&mdev->sdev->kobj, &mdev->attr_group);
> 
> We now have a function you should use instead,
> device_create_with_groups() that solves the race condition you just
> caused here by creating and notifying userspace that the device is
> present, _before_ creating the sysfs files for it.
> 

We are using device_create_with_groups(..) now and it works nicely here.

> 
> 
> > +	if (rc) {
> > +		dev_err(&pdev->dev, "sysfs_create_group failed rc %d\n", rc);
> > +		goto destroy_device;
> > +	}
> > +	dev_info(&pdev->dev, "Probe successful for %s\n", mdev->name);
> 
> Useless noise, remove it.
> 

Agreed, will remove.

> > +	return 0;
> > +destroy_device:
> > +	device_destroy(g_mic.mic_class, MKDEV(MAJOR(g_mic.mdev_id), mdev->id));
> > +unmap_aper:
> > +	iounmap(mdev->mmio.va);
> > +unmap_mmio:
> > +	iounmap(mdev->aper.va);
> > +release_regions:
> > +	pci_release_regions(pdev);
> > +disable_device:
> > +	pci_disable_device(pdev);
> > +free_device:
> > +	kfree(mdev);
> > +dec_num_dev:
> > +	g_mic.next_id--;
> > +	dev_err(&pdev->dev, "Probe failed rc %d\n", rc);
> > +	return rc;
> > +}
> > +
> > +/**
> > + * mic_remove - Device Removal Routine
> > + * mic_remove is called by the PCI subsystem to alert the driver
> > + * that it should release a PCI device.
> > + *
> > + * @pdev: PCI device structure
> > + */
> > +static void mic_remove(struct pci_dev *pdev)
> > +{
> > +	struct mic_device *mdev;
> > +	int id;
> > +
> > +	mdev = pci_get_drvdata(pdev);
> > +	if (!mdev)
> > +		return;
> > +
> > +	id = mdev->id;
> > +
> > +	sysfs_remove_group(&mdev->sdev->kobj, &mdev->attr_group);
> 
> No need to do this as:
> 
> > +	device_destroy(g_mic.mic_class, MKDEV(MAJOR(g_mic.mdev_id), mdev->id));
> 
> Will do it for you if you make the changes above.
> 

Indeed, it cleans up correctly.

We will incorporate all your suggestions for patch 1 (including the
header file cleanup in the other thread) and post a rev3 once it has
seen some validation.

Thanks for the review!
Sudeep Dutt

> thanks,
> 
> greg k-h


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists